NoPeopleAllowed: The Three-Step Approach to Weakly Supervised Semantic Segmentation

TL;DR

This paper introduces a three-step weakly supervised semantic segmentation method that generates high-quality pseudo masks from image-level labels, effectively handling class imbalance and missing labels, achieving competitive results.

Contribution

The novel three-step approach improves pseudo mask quality and addresses data issues, advancing weakly supervised segmentation with only image-level annotations.

Findings

Achieves 37.34 mean IoU on test set

Placed 3rd at the LID Challenge

Effectively handles class imbalance and missing labels

Abstract

We propose a novel approach to weakly supervised semantic segmentation, which consists of three consecutive steps. The first two steps extract high-quality pseudo masks from image-level annotated data, which are then used to train a segmentation model on the third step. The presented approach also addresses two problems in the data: class imbalance and missing labels. Using only image-level annotations as supervision, our method is capable of segmenting various classes and complex objects. It achieves 37.34 mean IoU on the test set, placing 3rd at the LID Challenge in the task of weakly supervised semantic segmentation.